MUMBAI, India, April 17 -- Intellectual Property India has published a patent application (202641024501 A) filed by Nandha Engineering College, Erode, Tamil Nadu, on March 2, for 'temporal facial dynamics driven audio visual emotion recognition using vision transformer.'
Inventor(s) include Praveen Kumar R; Sowmiya Devi P; Siva Harlne S; and Sneha R.
The application for the patent was published on April 17, under issue no. 16/2026.
According to the abstract released by the Intellectual Property India: "The present invention discloses an Al-driven accident detection and tral'lic risk alert system designed to accurately identify road accidents and accident-prone situations li"om real-time CCTV video streams using deep learning and multimodal reasoning techniques. The system analyzes visual tralfic data to detect collision events, assess severity levels, and identil)' unsalc driving behaviors. Unlike conventional static or rule-based surveillance approaches, the proposed frameiVork integrates contextual reasoning and risk prediction to enhance reliability and responsiveness. The invention employs a vision encoding architecture compnsmg a Segment Anything ivlodel (SAM) to extract high-resolution spatial representations of vehicles, road infrastructure, and trarlic interactions. These features are refined using an attention mechanism including a Convolutional Block Attention Module (CBAM) to emphasize accident-relevant regions and suppress background noise. The relined visual representations are processed through a Large Language Model (LLM) backbone to perform contextual reasoning, accident classification, and descriptive interpretation oftraffic incidents. A dedicated risk analysis module is incorporated to identify accident-prone behaviors such as sudden lane changes, over-speeding, congestion buildup, and road hazards. Risk activations are aggregated over time to generate explainable outputs including bounding boxes, directional indicators, and density-based heatmaps that highlight high-risk zones within monitored areas. The system further includes an automated alert mechanism that transmits accident location and severity information to emergency authorities through cloud-based or GSM communication modules. Experimental evaluation demonstrates that the proposed invention achieves improved detection accuracy, reduced false-positive rates, and stable real-time performance compared to traditional static and unimodal accident detection systems. The disclosed system is suitable for smart city traffic monitoring, emergency response coordination, and intelligent transportation systems, providing a scalable, proactive, and explainable solution for modern road safety management."
Disclaimer: Curated by HT Syndication.
